© ESO, 2011

1. Introduction

Massive stars have played an important role in galaxy evolution throughout cosmological time via their intense winds, ultraviolet radiation fields, chemical processing, and explosions. For instance, they dominate the rest-frame ultraviolet spectra of star-forming, Lyman-break galaxies, in which multiple “starburst” components (or mergers) are seen out to redshifts of at least z  ~  5 (Douglas et al. 2010). Indeed, they are thought to have been a major factor in the reionization of the early universe (Haiman & Loeb 1997), comprising the dominant component in the earliest galaxies, in which the star-formation rate appears to increase by a factor of ten over a period of just 200 Myr (Bouwens et al. 2011).

Population synthesis codes such as Starburst99 (Leitherer et al. 1999, 2010) form the bridge between our understanding of the physics and evolution of individual stars, and efforts to analyse entire populations on galaxy scales via interpretation of their integrated light. The sensitivity of the 8–10 m class telescopes has enabled spectroscopy of individual massive stars in galaxies at Mpc distances (e.g., Gieren et al. 2005; Kudritzki et al. 2008), including some in low-metallicity dwarf galaxies, in which the local conditions are close to those found in the early universe (e.g., Bresolin et al. 2006, 2007; Evans et al. 2007).

Unfortunately, we are currently limited to observations of the brightest stars in such galaxies. Only in the Galaxy and Magellanic Clouds can we potentially assemble large observational samples of massive stars that span a wide range of metallicity and intrinsic luminosities. Ultraviolet satellites have enabled substantial surveys of terminal wind velocities (e.g., Howarth & Prinja 1989; Prinja et al. 1990) and rotational velocities (Penny 1996; Howarth et al. 1997; Penny & Gies 2009). However, both observational and computational challenges have precluded the atmospheric analysis of a large, coherent sample of O-type stars from optical spectroscopy, with studies limited to several tens of objects at most (e.g. Herrero et al. 2000, 2002; Repolust et al. 2004; Mokiem et al. 2005, 2006, 2007; Massey et al. 2009). At lower masses, while large observational samples of early B-type stars are available, recent results have highlighted some serious problems regarding our understanding of massive-star evolution (Hunter et al. 2008, 2009; Brott et al. 2011). This leaves us in a situation where fundamental questions concerning the formation, evolution, and eventual fate of massive stars remain unanswered, particularly in the context of results that point to the majority of high-mass stars being in binary/multiple systems (e.g. Sana & Evans 2011).

The Tarantula Nebula (NGC 2070, 30 Doradus – hereafter “30 Dor”) in the Large Magellanic Cloud (LMC) is the brightest H ii region in the Local Group. It is comprised of multiple generations of star formation, with at least five distinct populations (Walborn & Blades 1997), similar to cluster complexes seen well beyond the Local Group (e.g., Bastian et al. 2006). At the heart of 30 Dor is Radcliffe 136 (R136, Feast et al. 1960), a massive, young stellar cluster (1–2 Myr; e.g., de Koter et al. 1998; Massey & Hunter 1998) dominated by early O-type and hydrogen-rich WN-type stars, some of which are thought to have current masses in excess of 150 M_⊙ (Crowther et al. 2010). To the north and west of R136 there is significant molecular gas (Werner et al. 1978; Johansson et al. 1998) with embedded massive stars, which has been suggested as a second generation of (perhaps triggered) star formation around R136 (Walborn & Blades 1997; Walborn et al. 1999a, 2002).

By virtue of its location in the LMC, the distance to 30 Dor is well constrained (we adopt a distance modulus to the LMC of 18.5, e.g., Gibson 2000) and the foreground extinction is relatively low compared to some of the most massive Galactic clusters. Moreover, the metallicity of the LMC (~50% solar) is well matched to typical results for high-redshift galaxies (e.g. Erb et al. 2006; Nesvadba et al. 2008).

Thus, 30 Dor provides an excellent opportunity to study a broad age range of massive stars within a single complex of star formation. In addition to the rich populations of O- and early B-type stars, there are over twenty Wolf-Rayet (W-R) stars, (including examples of both WN and WC subclasses, e.g., Breysacher et al. 1999), several transition Of/WN stars (Crowther & Walborn, in prep.), and candidates for high-mass young stellar objects (YSOs) identified from mid-infrared observations with the Spitzer Space Telescope (see Sect. 5). With its rich stellar populations, 30 Dor is the ideal laboratory in which to investigate important outstanding questions regarding the physical properties, binary fraction, chemical enrichment, and evolution of the most massive stars, as well as the relation of massive stars to the formation and evolution of stellar clusters (e.g., de Koter et al. 2008; Portegies Zwart et al. 2010).

Building on experience from the VLT-FLAMES Survey of Massive Stars (Evans et al. 2005, 2006), we introduce the VLT-FLAMES Tarantula Survey (VFTS), a multi-epoch spectral survey of over 800 massive stars in the 30 Dor region, including  ~ 300 O-type stars. In a series of articles, the VFTS data will be used to investigate the properties of stellar winds and rotational mixing in O-type stars, and to extend studies of rotational mixing in B-type stars. In particular, the surface abundances of O-type stars are expected to be modified via the effects of rotational mixing but, to date, nitrogen abundances have not been determined for a large observational sample. The inclusion of new models of N III in the fastwind model atmosphere code (Puls et al. 2005, 2011), combined with analytical methods such as those developed by Mokiem et al. (2005) or the use of a large model grid, means that such a critical test of evolutionary prodictions is now feasible.

30 Dor is known to have a rich binary population (Bosch et al. 2009) – a key feature of the VFTS is a multi-epoch observational strategy to obtain clear indications of binarity¹ in the large majority of the targets. This will add a valuable component to quantitative analysis of the spectra and interpretation of the results, enabling tests of the predictions of evolutionary models that include all of the relevant physical processes for both single stars and binary systems (e.g. Langer et al. 2008). The survey also includes multi-epoch spectroscopy in the inner part of 30 Dor with the FLAMES-ARGUS integral-field unit (IFU) to obtain an improved estimate of the velocity dispersion of single stars in R136, which will be used to determine the dynamical mass of the cluster.

This introductory article presents the significant observational material in one resource, comprising details of target selection, observational strategy and data reduction (Sect. 2), optical and infrared photometry of the targets (Sect. 3), spectral classifications for the observed massive emission-line and cooler-type stars (Sect. 4), and a discussion of the spectral properties of stars with published mid-infrared excesses (Sect. 5); concluding remarks are given in Sect. 6.

2. Spectroscopy

The Tarantula Survey employs three modes of the Fibre Large Array Multi-Element Spectrograph (FLAMES; Pasquini et al. 2002) instrument on the VLT:

• Medusa-Giraffe: The majority of the spectra were obtained usingthe Medusa fibre-feed to the Giraffe spectrograph. There are atotal of 132 Medusa fibres available for science (or sky)observations, deployable within a 25′ diameter field-of-view andwith a diameter of 12 on the sky.

• ARGUS-Giraffe: The ARGUS IFU was used to observe five pointings of R136 and its environs in the central part of 30 Dor. The IFU was used in the 052 per spatial pixel (spaxel) mode, such that its 22  ×  14 microlenses provide a total field-of-view for a single pointing of 12′′  ×  7′′ on the sky.

• UVES: In parallel to the ARGUS observations, the fibre feed to the red arm of the Ultraviolet and Visual Echelle Spectrograph (UVES) was used to observe a small sample of stars in the inner part of 30 Dor at greater spectral resolving power than that delivered by Giraffe.

The target selection is discussed in Sect. 2.1, followed by a description of the reductions of the three spectroscopic components of the survey (Sects. 2.2–2.4).

2.1. Target selection

The Medusa fibre configurations were prepared using two sources. Within a radius of  ~ 60′′ of the core, targets were (with a couple of exceptions) taken from Brian Skiff’s reworking of the astrometry from the UBV catalogue of Selman et al. (1999)². Targets beyond this region were selected from preliminary reductions of B- and V-band observations (with the B/123 and V/89 filters, respectively) with the Wide-Field Imager (WFI) at the 2.2 m Max-Planck-Gesellschaft (MPG)/ESO telescope at La Silla (from observations by J. Alves, B. Vandame & Y. Beletsky; programme 076.C-0888). Four overlapping WFI fields were observed that cover approximately one square degree – our FLAMES observations primarily include sources from the first (northeast) WFI pointing, supplemented to the west by targets from the second. The median seeing in the reduced images is  ~ 08, although the conditions were non-photometric and stars brighter than approximately 14th magnitude were saturated.

Source detections in the WFI frames were performed using the daophot package (Stetson 1987) within iraf. The astrometry was calibrated using stars from the UCAC-2 catalogue (Zacharias et al. 2004), which has an astrometric precision to better than 01 (sufficient for the diameter of the Medusa fibres).

In a bid to avoid significant selection biases we did not employ any colour cuts on our input target lists. The only constraint imposed was a magnitude cut (before final photometric calibration) of V  ≤  17 mag, to ensure adequate signal-to-noise (S/N) in the resulting spectra. The Fibre Positioner Observation Support Software (FPOSS) was used with our combined “Selman-Skiff” and WFI catalogue to create nine Medusa configurations, hereafter referred to as fields “A” through to “I”. Exactly 1000 targets were observed with the Medusa-Giraffe mode in these nine configurations. The reductions and characteristics of these data are discussed below. After rejection of foreground stars (and a small number of others due to issues relating to data quality, see Sect. 2.2.4), we have Medusa (or UVES) observations of 893 unique targets.

A full catalogue of the survey targets is given in Table 5 in which they are RA-sorted and given a running number, which hereafter serves as an identifier of the form VFTS ###. Given its importance and rich stellar populations, 30 Dor has been the subject of numerous studies over the past decades – the final column of Table 5 provides a thorough (but not exhaustive) list of previous aliases/identifications for our targets.

FLAMES has a corrected field-of-view of 25′. This means that with one central telescope pointing we were able to observe stars in some of the local environs of 30 Dor as well as in the main body of the H II region. The distribution of the Medusa (and UVES) targets is shown in Fig. 1, overlaid on a section of the V-band WFI mosaic. The targets are primarily located in the main 30 Dor nebula (NGC 2070), but also include at least three separate associations: 1) Hodge 301, approximately 3′ to the northwest of R136 (Hodge 1988), which is somewhat older than the rest of 30 Dor (20 to 25 Myr, Grebel & Chu 2000); 2) NGC 2060 (aka 30 Dor B and LHA 120-N 157B), 65 to the southwest of R136, which is associated with a supernova remnant (e.g. Chu et al. 1992); 3) SL 639 (Shapley & Lindsay 1963), 75 to the southeast. The survey targets span a total diameter just in excess of 20′, with over 500 of them within a 5′ (≈ 75 pc) radius from the centre of R136. Distances to each target from R136 (specifically, R136-a1: α  =  5^h38^m42. ^s39, δ  =   − 69°06′0291, J2000.0) are given in the fifth and sixth columns of Table 5, in arcmin and pc, respectively.

2.2. VLT Medusa-Giraffe spectroscopy

2.2.1. Observational strategy

The nine Medusa fibre configurations (“A”–“I”) were observed using three of the standard Medusa-Giraffe settings (LR02, LR03, and HR15N). These provide intermediate-resolution spectroscopy of the lines commonly used in quantitative analysis of massive stars in the 3960–5070 Å region, combined with higher-resolution observations of the Hα line to provide a diagnostic of the stellar wind intensity.

The wavelength coverage and spectral resolution (Δλ, as determined by the mean full-width at half-maximum of arc lines in the wavelength calibration exposures) is summarised in Table 1. We also give the effective resolving power, R (i.e., λ/Δλ), for the central wavelength of each setting.

Most of the Medusa observations were obtained over the period 2008 October to 2009 February, with a seeing constraint of  ≤ 12. Pairs of exposures were taken in each observing block (OB). To obtain sufficient S/N for quantitative analysis of each target, three OBs at both the LR02 and LR03 settings were observed, and two at HR15N. No explicit time constraints were placed on the execution of these OBs and, in the majority of cases, the sequences at a given wavelength setting were observed consecutively (but not in all, e.g., the LR03 observations of Field D). The modified Julian Dates (MJD) for each OB are given in Tables A.1 and A.2 in the Appendix.

The key observational feature of the survey is three repeat OBs at the LR02 setting to detect radial velocity variables (epochs 4, 5, and 6 in the Appendix). These were scheduled in the service queue such that a minimum of 28 days had passed between execution of the third and fourth epochs, similarly between the fourth and fifth. The sixth epoch was obtained in 2009 October. The inclusion of the extra epoch a year later significantly helps with the detection of both intermediate- and long-period binaries. An important part of the interpretation of these multi-epoch spectra will be modelling of the detection sensitivities to put firm limits on the observed binary fraction (Sana et al., in prep.).

Owing to the nature of service-mode observations, a small number of the OBs at the LR02 wavelength setting were repeated for operational reasons, e.g., deterioration of the seeing beyond the required constraints during the exposure, failure of the tracking on the guide star, etc. In cases where full exposures (i.e. 1815 s) were completed, these observations were retained – even if the S/N is lower than in other frames, there will be useful radial velocity information for some targets. This leads to the “extra” frames for Fields A, B, E, and I in Tables A.1 and A.2. There is also an extra LR03 exposure of Field F (“LR03 01 [c]”), included out-of-sequence of its execution. This OB was aborted mid-execution because the telescope was being used for interferometric observations – the one completed exposure was retained because it provides additional radial velocity information, for example, in the case of R139 (Taylor et al. 2011).

2.2.2. Reductions

The ESO Common Pipeline Library (CPL) FLAMES reduction routines (v2.8.7) were used for all of the initial data processing, i.e. bias subtraction, fibre location and (summed) extractions, division by a normalised flat-field, and wavelength calibration. The subsequent reduction stages were:

• Heliocentric correction: All spectra were corrected to theheliocentric frame, using the iraf packages rvcorrect anddopcor.

• Sky subtraction: Each sky fibre was inspected for signs of an on-sky source or cross-contamination from bright spectra/emission lines from adjacent fibres on the detector (see Sect. 2.2.4). Any contaminated sky fibres were rejected (typically one or two per frame) brefore creating a median sky spectrum, which was then subtracted from all fibres in that observation.

• Error spectrum: An error spectrum was produced by the pipeline for each fibre as part of the reduction process. For each wavelength bin it recorded the statistical error arising from the different stages in the reduction, e.g., bias level, detector gain, read-out noise. These data were combined with the errors on the median sky spectrum to obtain an estimate of the overall error for each spectrum.

• Cosmic-ray rejection: Our Medusa-Giraffe exposures were taken in consecutive pairs within the OBs. To clean the extracted spectra of cosmic rays we employed a technique developed (by I.D.H.) for the 2dF survey of the Small Magellanic Cloud (SMC) by Evans et al. (2004). For each spectrum, the ratio of the two exposures was calculated. A boxcar 4-sigma clip over 100 wavelength bins was then performed on this ratio value. Any unexpected and significant deviations in the ratio are indicative of a feature in only one of the observations. Because the exposures were consecutive, it is safe to assume that a transient feature was a cosmic ray. The pixels identified as suspect were rejected, then replaced with the value from the sister exposure, appropriately scaled by the ratio of the surrounding region. This approach results in good removal of cosmics, but is not perfect (which would require inter-comparison between three exposures).

• Rejection of foreground stars: Preliminary inspection of the LR02 data for each target was used to identify foreground cool-type stars to exclude them from our final catalogue (employing a velocity threshold of v_r  <  100 km   s^-1 for rejection). A total of 102 stars were excluded at this stage, which also included a small number of cool stars with LMC-like radial velocities but with very poor S/N.

As an example of the final S/N of the spectra we consider VFTS 553, one of the faintest Medusa targets (V  =  17.0 mag, for which the data reveals an early B-type spectrum). The mean S/N ratio in the individual exposures per resolution element were  ~ 50 for all three wavelength settings (determined using line-free regions of the stellar continuum). In contrast, the S/N for VFTS 527 (R139, one of the brightest targets) exceeds 400 per resolution element in the best spectra (Taylor et al. 2011).

2.2.3. Nebular contamination

Related to the sky subtraction, one of the principal limitations of fibre spectroscopy is the subtraction of local nebular emission. Indeed, at the distance of the Magellanic Clouds, even (seeing-limited) long-slit spectroscopy can suffer difficulties from spatially-varying nebular emission (cf. the HST spectroscopy from, e.g. Walborn et al. 2002). Given the significant nebular emission in 30 Dor, the majority of the Giraffe spectra have some degree of nebular contamination.

The combination of extended nebular emission with the stellar flux from a point-source means that, because the seeing conditions vary for different epochs, the relative intensity of the nebular contamination can vary. Apart from a minority of spectra with particularly strong emission, this is not a big problem for the hydrogen Balmer lines, where the core nebular profile can (effectively) be ignored in quantitative analysis (e.g. Mokiem et al. 2006). However, care is required when selecting He I lines for analysis (most notably λ4471) because small changes in the nebulosity could be interpreted as evidence for binarity if unchecked with other lines.

2.2.4. Cross-contamination of spectra

To keep the observing as simple and homogeneous as possible, a constant exposure time was used for each fibre configuration. A consequence was that bright stars can cross-contaminate adjacent spectra of fainter stars on the detector after dispersion. This cross-contamination between fibres was generally minimal, but inspection of the reduced spectra revealed a small number of fibres that were contaminated by bright emission lines in W-R or “slash” stars and, in a couple of cases, by luminous supergiants with large continuum fluxes.

Thus, for the observation obtained in the best seeing, we inspected the spectra adjacent to all “bright” stars (with counts greater than 10 000) for each Medusa field and wavelength setting. In some instances the contaminated spectra were sky fibres (rejected as described above), but five stellar targets were omitted because they were pathologically contaminated by very strong emission lines. A further 22 stars have some element of cross-contamination – typically an artificial broad emission “bump” at λ4686 from strong He II in emission-line stars in adjacent fibres. These are noted in the final column of Table 5 – quantitative spectral analysis of these stars may not be possible, but the radial velocities from other regions/lines will still be useful to investigate binarity, gas dynamics, etc.

2.3. VLT ARGUS-Giraffe spectroscopy

The Medusa data were complemented by five pointings within the central arcminute of 30 Dor with the ARGUS IFU, as shown in the main panel of Fig. 2. The coarser ARGUS sampling was used (052/microlens, giving a field-of-view of ), with a seeing constraint of  ≤ 08.

These regions are densely populated with stars, particularly in the core of R136. The first IFU pointing (“A1” in the figure) was located on the core, with three pointings immediately adjacent. The fifth pointing was located to the NNE of R136 to target a reasonable number of stars at a slightly larger radius from the core.

Full spectral coverage for quantitative analysis is best obtained with AO-corrected or HST spectroscopy – our intention here was to probe the dynamics of these inner regions, again with follow-up observations to identify binary systems. Thus, only the LR02 Giraffe setting was used. The resulting wavelength coverage was comparable to that from the Medusa observations, but at greater resolving power because of the smaller aperture (see Table 1).

Two OBs were observed without time restrictions and, similar to our strategy for the Medusa data, follow-up epochs (third and fourth) were observed with a minimum interval of 28 days. All these data were obtained over the period 2008 October to 2009 March, with a final (fifth) epoch observed in 2009 December/2010 January.

As with the Medusa observations, a number of the ARGUS OBs were re-observed owing to changes in conditions and other operational issues. The full list of completed exposures for the ARGUS frames is given in Table A.3. These epochs also apply to the UVES observations (Sect. 2.4), which were taken in parallel to the ARGUS data.

The seeing conditions are more critical to these IFU observations than for the Medusa frames, so the adopted nomenclature is slightly different for repeated OBs. The exposures obtained under the best conditions (ascertained from the seeing value recorded in the file headers at the time of observation) are identified as the “a + b” pair, with other exposures following in order of execution (as “c + d” etc.), e.g., epoch four for the fourth pointing (“A4”).

2.3.1. Reduction

The ARGUS frames were reduced using the same methods as for the Medusa data apart from the sky subtraction and the combination of spectra from adjacent spaxels on the sky (individual stars typically extend over several spaxels). The extracted (cosmic-clipped) spectra were corrected to the heliocentric frame, then combined as follows.

Sources were selected if they appeared isolated in the reduced IFU datacube and if they could be matched to a star (or multiple, densely-packed stars) in an archival HST F555W image (from the early release science observations of 30 Dor taken with the Wide-Field Camera Three, WFC3). Less isolated sources were also extracted if their spectra could be distinguished from those of their surroundings, and if they had a matching bright source in the WFC3 image. (Some sources are obviously multiple in the WFC3 image but their spectra were retained because they might still prove useful in the analysis of the velocity dispersion of R136.) The spaxels combined for a given source were selected on the basis that they showed the same spectral features (and relative strengths) as the brighest/central spaxel of that source. There are small positional shifts of approximately one spaxel between the different epochs of a given pointing. These shifts were taken into account when defining (for each frame) the spaxels that contribute to each source.

Spectra were extracted for a total of 41 sources from the ARGUS frames. The 37 unique ARGUS sources are appended to the end of Table 5, with coordinates from centroids of matching sources in the WFC3 image (transformed to the same astrometric system as the Selman–Skiff catalogue for consistency). To distinguish from targets observed with Medusa and/or UVES, these sources are given a separate series of RA-sorted identifiers, starting with VFTS 1001. Photometry is available for all but three from Selman et al. (1999), as listed in Table 5. Note that four Medusa/UVES sources were also observed with ARGUS (VFTS 542, 545, 570, and 585), as noted in the final column of Table 5. The five ARGUS pointings and the location of the extracted sources are shown in Fig. 2.

The four or five spaxels with the lowest counts in each pointing were used for local sky spectra. Even so, for most sources the nebular subtraction is still far from perfect given the small-scale variations discussed earlier. Before combining the sky spectra, a 5σ-clip (compared to the noise of the median spectrum) was employed to remove remaining cosmic rays or artefacts. A weighted-average sky spectrum was created, then subtracted from the spectra from each of the source spaxels.

Differential atmospheric refraction means that the apparent slope of the continuum of a given source varies from spaxel to spaxel. Thus, each spectrum was normalised individually before combining the contribution of different spaxels. This normalisation was done by a spline-fit across carefully selected continuum regions, then division of the spectrum by the resulting smooth curve. This method generally gave an excellent fit to the continuum, but is less certain for spectra with broad emission lines, for which the continuum is hard to define. The final spectrum of each source is a weighted average of the normalised spectra from the individual spaxels, again employing a 5σ-clip around the median to remove spurious pixels.

2.4. VLT FLAMES-UVES spectroscopy

The fibre feed to the red arm of UVES was used to observe 25 stars in parallel to the ARGUS observations. Twenty of these were also observed with Giraffe – our aim was to exploit the availability of UVES to obtain broader spectral coverage and additional epochs for radial velocity measurements (at even greater precision). The UVES targets are indicated by “U” in the second column of Table 5, and are all from the Selman et al. (1999) catalogue, i.e., in the inner part of 30 Dor near R136.

2.4.1. Reduction

The two CCDs in the red arm of UVES were processed separately. With the λ5200 central wavelength setting, the short-wavelength CCD provided coverage of 4175 to 5155 Å, with the long-wavelength CCD spanning 5240 to 6200 Å. The UVES CPL routines (v4.3.0) from ESO were used for the preliminary reduction stages: bias correction, fibre extraction, wavelength calibration, and division by a normalised flat-field frame. Our own idl routines were then used for merging of the orders and velocity correction to the heliocentric frame. For these bright targets and at this resolving power, the sky background was relatively low – given the problems relating to merging of the echelle orders, subtraction of the sky spectra resulted in a poorer final data product, so they were not used. From analysis of the reduced arc calibration frames, the delivered resolving power was R  ~  53 000.

3. Photometry

The approximate photometric calibration of the WFI imaging was sufficient for target selection (Sect. 2.1), but quantitative spectroscopic analysis requires precise photometry. Given the complexity of 30 Dor (in terms of source detection in regions of bright nebulosity) and the non-photometric conditions/saturation of bright stars in the WFI observations, we revisited the data with a more refined photometric analysis.

From combining the Selman-Skiff catalogue and our analysis of the WFI frames we obtained B- and V-band photometry for 717 of the VFTS targets. These were then supplemented by photometry for 68 stars from Parker (1993) and for 91 stars from new imaging at the Cerro Tololo Inter-American Observatory (CTIO), Chile. Each of these photometric sources is discussed briefly below.

Moreover, when undertaking quantitative spectral analysis of massive stars in regions of variable extinction, near-IR photometry can be a useful input towards the determination of accurate stellar luminosities (e.g. Crowther et al. 2010). Cross-matches of the VFTS targets with published near-IR data are discussed in Sects. 3.6 and 3.7.

We note that there is a significant amount of imaging of selected regions of 30 Dor at better spatial resolution, principally from the HST. Our philosophy here is to compile optical photometry from seeing-limited, ground-based images, which are reasonably matched to the aperture of the Medusa fibres. Thus, even if our “stars” are actually multiple objects, the light received by the fibre is comparable to that measured from the adopted imaging.

3.1. “Selman” photometry

The data used for target selection in the central region were taken from the Selman-Skiff catalogue, which was obtained using the Superb-Seeing Imager (SUSI) on the 3.5 m New Technology Telescope (NTT) at La Silla, Chile. The conditions were photometric with sub-arcsecond seeing, and short exposures (10–30 s) were used to avoid saturation. The fine pixel scale of SUSI (013/pixel) means that crowding is less problematic than in the WFI frames. We therefore adopt the B- and V-band photometry from Selman et al. (1999) for the 167 stars observed from their catalogue, listed with the reference “S” in Table 5. To illustrate the selection effects employed on the FLAMES sources (i.e. V  ≤  17 mag and no colour constraint), a colour–magnitude diagram for the 167 stars with photometry from Selman et al. is shown in Fig. 3; their other sources (V  ≤  18 mag) are also shown.

3.2. WFI photometry

Our primary source of photometry is from point-spread function (PSF) fitting of sources in the WFI frames with the stand-alone version of daophot. We are mainly interested in the relatively bright population so we adopted a detection threshold of 20 times the root-mean-square noise of the local background to avoid spurious detections. We then applied a cut of 0.2  <  sharpness  <  1 to further filter those detections for extended objects, etc. The resulting B- and V-band catalogues were merged together using the daomatch and daomaster routines (Stetson 1994), i.e. the final catalogue includes only those sources detected in both bands.

Zero-points for the WFI photometry were determined with reference to 67 stars (within a search radius of less than 04 and V  ≤  18.0 mag) that overlap with the Selman–Skiff catalogue. The resulting residuals for these matched sources are shown in Fig. 4, with standard deviations of  ≤ 0.1 mag in both bands. The second WFI pointing was bootstrapped from the first using the mean differences between overlapping stars.

Photometry of 550 stars is adopted from analysis of the WFI images (calibrated using the catalogue from Selman et al. 1999). These are listed with the reference “W” in Table 5. Their colour–magnitude diagram is shown in Fig. 5; also shown are the  ~ 2000 other WFI sources (V  ≤  18 mag) within a radius of 12′ from R136. There are ten sources with WFI photometry in Table 5 with V  ≤  14.5 mag, each of which agrees with the CTIO photometry (Sect. 3.5) to within 0.1 mag, i.e., within the dispersion of the calibrations in Fig. 4.

3.3. “Parker” photometry

To supplement the WFI and Selman photometry we first turned to the catalogue from Parker (1993) in 30 Dor, using the reworked astrometry from Brian Skiff³. A comparison between the Parker results and the calibrated WFI photometry for 219 matched stars (within a radius of 05) yielded only small residuals of ΔV  =  0.01 (σ  =  0.2) and ΔB  =   − 0.03 (σ  =  0.2) mag, as shown in Fig. 6.

A radial search (of 05) and then visual matching of the VFTS targets without WFI or Selman photometry yielded 68 matches in the Parker catalogue (with a median radial offset of 011). For our current purposes, we adopt his photometry for these 68 stars, listed with the reference “P” in Table 5.

3.4. MCPS photometry

As a potential source of photometric information for the remaining 108 stars we turned to the LMC catalogue from the Magellanic Clouds Photometric Survey (MCPS; Zaritsky et al. 2004), which includes photometry of some of the bright sources from Massey (2002). However, we found large uncertainties on the MCPS photometry in the 30 Dor region.

We first considered cross-matched sources from the MCPS catalogue as an external check on the photometric calibration of the WFI frames. Adopting a search radius of less than 10, a total of 5726 “matches” were found between the northeastern WFI frame and the MCPS. The residuals for B and V (in the sense of WFI  −  MCPS) are shown in Fig. 7 as a function of MCPS colour. The standard deviation of the residuals in both bands is  ~ 0.25 mag, not unreasonable considering the potential for specious matches and complications relating to blending and nebulosity in this field – the typical seeing from the MCPS is 15 (with 07 pixels), but extended up to 25 (Zaritsky et al. 2004), while the photometry from Massey (2002) used an aperture of 162. Although there is reasonable agreement in the zero-point calibrations between the WFI and MCPS data, the “plume” to brighter MCPS magnitudes is particularly notable in Fig. 7.

Concerned by the potential of inter-CCD calibration problems in the WFI data, we investigated the distribution of the residuals as a function of both declination and right ascension, as shown in Figs. 8 and 9, respectively. The residuals between the WFI and MCPS photometry are noticeably larger over the main body of 30 Dor (centred at α  =  5. ^h645, δ  =   − 69 .°101, and with the densest nebulosity having a diameter of  ~ 6′). Larger residuals can also be seen in the region of α  =  5. ^h702, the centre of the dense cluster NGC 2100. The residuals are predominantly in the sense of brighter magnitudes from MCPS compared to the WFI frames, which suggests that they primarily arise from unresolved blends or nebular contamination.

3.5. CTIO photometry

Thus, to obtain optical photometry for the majority of the remaining FLAMES targets, we observed 30 Dor on 2010 December 27 with the Y4KCAM camera on the CTIO 1 m telescope, operated by the SMARTS consortium⁴. The camera is equipped with an STA 4064  ×  4064 CCD with 15 μm pixels, yielding a scale of 029 pixel^-1 and total field-of-view of 20′  ×  20′ at the Cassegrain focus.

Observations were obtained in photometric conditions with good seeing (less than 13), using B- and V-band filters in the Kitt Peak system⁵. A range of exposure times in each band were used to avoid saturating bright stars (from 10 s up to a maximum of 400 and 300 s in the B- and V-bands, respectively). Instrumental magnitudes were obtained using PSF-fitting routines in daophot, then observations of standard stars taken on the same night in Selected Area 98 (Landolt 1992) were used to transform the photometry to the Johnson-Kron-Cousins system and to correct for atmospheric extinction. The standard observations spanned a range of airmass (from 1.05 to 2.60) and included stars over a broad range of colours (− 0.3  ≤  (B − V)  ≤  1.7 mag).

Robust matches were found for 91 of the remaining FLAMES targets, with photometry from the CTIO imaging listed in Table 5 with the reference “C”. Given the independent calibration using published standards, we do not attempt to transform the photometry of these stars onto the same exact system as the WFI photometry. However, they agree reasonably well – a comparison of nearly 800 matched stars yields mean residuals of ΔV and ΔB  ≤  0.05 mag (with σ  ~  0.2 mag in both bands).

Seven of the remaining targets were beyond the western and southern extent of the CTIO images (VFTS 002, 003, 014, 016, 017, 739, and 764) and five did not have counterparts in the CTIO catalogue owing to nearby blends or the PSF-fitting criteria (VFTS 092, 172, 301, 776, and 835). Photometry is available from the MCPS catalogue for four of these (VFTS 016, 092, 739, 764), each of which is over 8′ from the core of 30 Dor, so the problems discussed in Sect. 3.4 should be minimised.

The last five targets without optical photometry (VFTS 145, 147, 150, 151, & 153) are in the dense “Brey 73 complex”, which was first resolved by Testor et al. (1988) and later observed with the HST by Walborn et al. (1995, 1999b). The HST photometry from Walborn et al. (1999b) is adopted in Table 5 for the three (visually) single stars: VFTS 145, 150, & 153. The two brightest members (#1 and #2 from Testor et al., VFTS 147 and 151) were resolved into separate bright components by the HST imaging, i.e. the Medusa fibres will contain contributions from these, and any future analysis will have to consider their relative fluxes (and colours).

3.6. Near-IR photometry

The extensive near-IR imaging survey of the Magellanic Clouds from Kato et al. (2007)⁶ provides JHK_s photometry for nearly all of our FLAMES targets. The mean seeing at J, H, and K_s was 13, 12, and 11, respectively, i.e., well-matched to the on-sky Medusa fibre aperture.

To identify IRSF counterparts to the FLAMES targets we employed an astrometric search radius of 05, then overlaid the resulting list on the WFI V-band images to reject specious matches. The IRSF magnitudes (JHK_s, and their associated photometric errors) are given for the FLAMES targets in Table 6. The IRSF “quality” flag in the final column indicates the source detection in each of the three bands as follows (for futher details see Kato et al. 2007): “1”, point-like; “2”, extended source; “3”, saturated; “4”, faint; “5”, odd shaped (e.g. double sources); “0”, no detection.

The Two Micron All Sky Survey (2MASS, Skrutskie et al. 2006) catalogue was used to calibrate the IRSF astrometry and to provide checks on the photometry (see Kato et al. 2007). However, we note that the IRSF-SIRIUS filter-set was designed to match that of the Mauna Kea Observatories near-IR filters (Tokunaga et al. 2002), and is therefore slightly different to that used by 2MASS. Transformation equations between the IRSF-SIRIUS and 2MASS systems are given by Kato et al. (2007) and Kucinskas et al. (2008); in practise, the corrections between the two are relatively small, with typical differences of less than 0.05 mag for (J − K_s)_IRSF  <  1.7 mag (Kucinskas et al. 2008).

There are five FLAMES targets without good IRSF matches within 05: VFTS 275, 503, 620, 823, and 828. Additionally, VFTS 151 has two potential matches, but is excluded following the discussion of multiplicity in Sect. 3.5. In some instances there were two potential matches both within a radius of less than 03. In approximately half of these one set of the IRSF measurements was obtained in the “periphery” of the dither patterns (so the S/N is lower) and these values were omitted. However, for seven targets (VFTS 125, 330, 368, 374, 377, 383, 384) both IRSF detections are good spatial matches and not saturated. For these sources we compared the observational conditions of the relevant exposures (Table 2, Kato et al. 2007) and adopted the values obtained in the best seeing. As noted in Table 5, VFTS 240 appears slightly extended (or as a blended source) in the optical WFI image; the IRSF catalogue has a counterpart approximately 05 from the FLAMES position, with J  =  16.31  ±  0.06 mag, but it was not detected in the other two near-IR bands.

Similar comparisons were undertaken between the IRSF catalogue and the unique ARGUS targets (i.e. those numbered from 1001 to 1037, see Sect. 2.3). Good matches (all within 03) were found for 31 sources, as summarised in Table 6. Six ARGUS sources are without IRSF photometry: VFTS 1012, 1014, 1015, 1019, 1024, and 1025.

3.7. Cross-references with 2MASS

Near-IR photometry for some of our targets is also available from the 2MASS catalogue. However, with a pixel size of 20, its spatial resolution is inferior to the IRSF data. Nonetheless, we include cross-identifications to 2MASS for the VFTS targets in the final column of Table 5 for completeness. Our approach was similar to the comparison with the IRSF catalogue, with visual inspection of potential matches to reject those notably blended in the WFI data or with positions offset by contributions from other nearby stars/nebulosity. In particular, given the crowding in the central region around R136 and the limited angular resolution of 2MASS, we did not attempt cross-matching within the central 30′′. Only cross-matches with 2MASS photometric qualities of either “A” or “B” (i.e. S/N ≥ 7) were retained, leading to 2MASS identifications for 227 of our targets.

4. Spectral classification

Following reduction, the first epoch of LR02 and LR03 spectra of each target were inspected; approximately 300 targets display He II absorption, which is indicative of an O (or B0) spectral type. However, the multi-epoch nature of the VFTS spectroscopy complicates precise classification and entails significant analysis, which is being undertaken as part of studies towards binarity and the determination of stellar radial velocities (Sana et al.; and Dunstall et al., both in prep.). Detailed classifications of the O- and B-type spectra will therefore be given elsewhere. Here we present classifications for the two smaller spectral groups in the survey: the W-R/massive emission-line stars and the cooler stars (of A-type and later).

4.1. Wolf-Rayet and “slash” stars

Massive emission-line stars in the LMC have been well observed over the years, from the seminal Radcliffe Observatory study by Feast et al. (1960), to narrow-band imaging surveys to identify W-R stars (e.g. Azzopardi & Breysacher 1979, 1980), comprehensive spectroscopic catalogues (e.g. Breysacher 1981; Breysacher et al. 1999), and monitoring campaigns to study binarity (e.g. Moffat 1989; Schnurr et al. 2008).

The W-R and transitional “slash” stars (see Crowther & Walborn, in preparation) observed by the VFTS are summarised in Table 2 (and are highlighted in green in Fig. 1). In addition to those in R136 (Crowther et al. 2010), these stars comprise some of the most massive stars in 30 Dor and will provide valuable insights into some of the most critical phases of massive-star evolution.

Previous spectral types are included in Table 2, with revised classifications in the final column if new features and/or evidence for companions are present in the FLAMES data. The new data reveal massive companions in at least two of these objects: VFTS 402 and 509 (BAT99-95 and 103, respectively). Quantitative analysis of these data is now underway. Note that VFTS 527 (aka R139), has been treated as a W-R star by some past studies; the VFTS observations have revealed it as an evolved, massive binary system comprised of two O Iaf supergiants (Taylor et al. 2011).

4.2. Discovery of a new W-R star in the LMC

The combined FLAMES spectrum of VFTS 682 is shown in Fig. 10. Classified as WN5h (in which the “h” suffix denotes the presence of hydrogen lines), this is a previously unknown W-R star. From qualitative comparisons of the individual spectra there is no evidence for significant (Δv  ≳  10 km   s^-1) radial velocity variations.

In their analysis of the Spitzer-SAGE survey of the LMC (Meixner et al. 2006), VFTS 682 is included by Gruendl & Chu (2009, hereafter GC09) in their list of “definite” young stellar objects (YSOs). Near-IR (from the IRSF catalogue) and Spitzer photometry for VFTS 682 are summarised in Table 4. Near-IR photometry is also available from Hyland et al. (1992, infrared source 153) and 2MASS. The evolutionary status of VFTS 682 and the origins of its mid-IR excess are discussed by Bestenlehner et al. (in prep.).

4.2.1. A new B[e]-type star adjacent to R136

The ARGUS spectrum of VFTS 1003 (Fig. 11) is particularly striking owing to a large number of Fe II emission lines. It is similar to that of GG Carinae (see Walborn & Fitzpatrick 2000) but with emission lines from [Fe II]; some weak He I absorption lines are also present. There is forbidden [S II] and [O III] emission at λ4069 and λ4363, respectively, although from the present data it is not clear if these are from the local nebulosity or are intrinsic to the star. There do not appear to be any significant radial velocity shifts between the individual spectra. Its near-IR colours (J − H)  =  0.54 and (H − K_s)  =  1.43 mag (from the IRSF catalogue) place it in a comparable region to the B[e] stars from Gummersbach et al. (1995). We also note that it is a single, isolated source in the high angular-resolution near-IR images from the Multiconjugate Adaptive-optics Demonstrator (MAD) from Campbell et al. (2010).

Lamers et al. (1998) presented a classification framework for B[e]-type stars, with GG Car classified as a B[e] supergiant. The spectrum of VFTS 1003 warrants a B[e] classification, but does not allow us to distinguish between an evolved B[e] supergiant and a pre-main sequence (“Herbig”) B[e] star (cf. the criteria from Lamers et al.). Nevertheless, the presence of such a rare object a mere 85 from the core of R136 certainly warrants further study. In particular, photometric monitoring would help to distinguish between the two evolutionary scenarios: small variations (~0.2 mag) would be expected for a B[e]-type supergiant, whereas much larger and irregular variations (related to accretion processes) would be expected for a “Herbig” object (Lamers et al. 1998).

4.3. Classification of later-type stars

There are 91 stars with classifications of A-type or later that were retained as likely members of the LMC. For the purposes of classification these were assumed to be single stars, i.e. all of the available LR02 and LR03 spectra were stacked and co-added; classifications are given in Table 3.

The small number of A-type spectra were classified with reference to the standards presented by Evans & Howarth (2003) and the metal-poor A-type stars from Evans et al. (2006). The new FLAMES data do not include the Ca K line, so the primary temperature diagnostic is the intensity of the metal lines; luminosity types were assigned on the basis of the Hγ equivalent-width criteria from Evans et al. (2004).

Except for a small number of G-type stars (that were classified following the criteria from Evans & Howarth 2003) we employ broad classification bins for the cooler types. These are not a key component of our scientific motivations and due to their red colours have relatively low S/N ratios. The spectral bins adopted in Table 3 encompass a range of types: “Early G” (G0-G5); “Late G/Early K” (G5-K3); “Mid-late K” (K3-M0); and “Early M”. Three carbon stars with strong C₂ Swan bands are also included in the sample.

We do not attempt luminosity classifications for stars of F-type and later. On the basis that their radial velocities are consistent with membership of the LMC, from distance arguments they are notionally supergiants or bright giants (i.e. classes I and II). Indeed, five stars in Table 5 are sufficiently bright⁷ to be considered “red supergiants”: VFTS 081, 198, 236, 289, and 793, each of which is encircled in red in Fig. 1.

5. Spitzer YSO candidates

Prompted by the mid-IR behaviour of VFTS 682 (Sect. 4.2), we cross-matched the survey targets with the catalogues of “definite” and “probable” YSOs from GC09 to investigate their spectral properties.

The SAGE data comprise imaging with two Spitzer instruments: the InfraRed Array Camera (IRAC) at 3.6, 4.5, 5.8, and 8.0 μm, and the Multiband Imaging Photometer for Spitzer (MIPS) at 24, 70, and 160 μm. The angular resolution of the IRAC images ranges from 17 to 20, with coarser resolution of 6′′, 18′′, and 40′′, at 24, 70, and 160 μm, respectively (Meixner et al. 2006). To compare our targets with the GC09 catalogues, we employed a conservative search radius (compared to the IRAC resolution) of r  <  25, which yielded five potential matches in the “definite” YSO list, and 11 potential matches in the “probable” YSO list. Each potential match was then examined using the optical WFI images and, where possible, the MAD near-IR imaging from Campbell et al. (2010), and the imaging from the High Acuity Wide-field K-band Imager (HAWK-I) which was used to calibrate the MAD data.

The near-IR images are particularly helpful to identify robust counterparts. A notable false match at a distance of 12 from VFTS 476 is GC09 053839.69 − 690538.1. The optical WFI image reveals only the VFTS target whereas, as shown by Fig. 13 from Campbell et al. (2010), there is a very red source at the GC09 position, with VFTS 476 the object to the south⁸. Near- and mid-IR photometry of matched sources from GC09 are summarised in Table 4, each of these is now discussed in turn.

VFTS 016: This is the massive “runaway” star from Evans et al. (2010, “30 Dor 016”, classified as O2 III-If^∗, which features in the list of probable YSOs from GC09.

Mid-IR observations have been used by, e.g., Gvaramadze & Bomans (2008) and Gvaramadze et al. (2010, 2011) to identify bow shocks associated with runaway stars. For instance, the SAGE 24 μm images were used to investigate six O2-type stars in the LMC thought to be runaways, including VFTS 016 (Gvaramadze et al. 2010). Of the six candidate runaways, only BI 237 was reported to have a bow shock from inspection of the MIPS images, consistent with the expectation from Gvaramadze & Bomans (2008) that approximately 20% of runaways have an associated bow shock.

Although the Spitzer resolution at 24 μm is less than ideal in a region as crowded as 30 Dor, VFTS 016 is relatively isolated at a projected distance of 120 pc from the core of R136. The 24 μm magnitude from GC09 for VFTS 016 (see Table 4) suggests a strong mid-IR excess – perhaps associated with a bow shock (but not extended sufficiently to be detected by Gvaramadze et al.).

VFTS 178: A visually bright star, the spectrum of VFTS 178 appears as a relatively unremarkable supergiant, classified O9.7 Iab (see Fig. 12). The star was previously classified as B0.5 I by Schild & Testor (1992), with the new high-quality data revealing sufficiently strong He II λ4542 absorption that a slightly earlier type is required. No obvious radial velocity variations are apparent from inspection of the FLAMES data.

VFTS 320: This was the closest match to GC09 053821.10 − 690617.2 (with VFTS 316 at a distance of only 12). As noted in Table 5, the spectrum of VFTS 320 suffers from contamination from an adjacent fibre on the detector (see Sect. 2.2.2), manifested by a broad emission bump (approximately 4% above the continuum) at λ4686. Aside from this wavelength region, the spectrum is that of an early B-type star with considerable nebular contamination plus weak Fe II emission, which resembles that from a Be-type star. The adjacent object on the detector was VFTS 147, classified in Sect. 4.1 as WN6(h). This does not display Fe II emission (indeed, the He II λ4686 emission is by far the strongest line), i.e., the Fe II features seen in VFTS 320 are genuine. The spectrum (with the λ4686 region omitted) is shown in Fig. 12.

VFTS 345: A single object in the WFI and HAWK-I frames, the co-added spectrum of VFTS 345 is shown in Fig. 12 and is classified as O9.7 III(n) (cf. the published classification of B0 V from Bosch et al. 1999).

VFTS 410: This target is P93-409 (Parker 1993), located in “Knot 3” to the west of R136 (Walborn 1991). Classified as O3-6 V by Walborn & Blades (1997), HST imaging and spectroscopy by Walborn et al. (1999a, 2002) resolved two bright components with spectral types of O8.5 V and O9 V. A third, much redder (visually fainter) source was reported just 04 northwest of the two massive stars by Walborn et al. (1999a), as well as several very compact nebulosities to the northeast. This complex would be resolved poorly by the Spitzer observations, but is clearly an active site of star formation.

For completeness, we show the combined spectrum of VFTS 410 in Fig. 12; there is significant nebular emission superimposed on the stellar profiles. Given the knowledge that there are two luminous components contributing to these data, it is worth noting that no radial velocity variations are seen from a comparison of the available spectra. Even if these are not in a bound binary system, their proximity in such a star-formation complex suggests they are associated. This object highlights the benefit of high-resolution imaging for some systems (from HST, MAD, etc.) when attempting to interpret the new spectroscopy.

VFTS 464: This is the star at the centre of the impressive bow-shock feature discovered by Campbell et al. (2010) from the MAD images. In Fig. 13 we show new combined colour images from MAD (with different intensity scalings). There is a source 04 from the central object that is approximately three times fainter – given the 12 aperture of the Medusa fibres, the spectra of the central star are likely contaminated by the companion.

As one might expect, the FLAMES spectra are heavily contaminated by nebular emission, which appears to include forbidden lines such as [S II] λλ4069-76, [Fe III] λ4702, and [Ar IV] λ4711. The combined blue-region spectrum is shown in Fig. 12 – note the weak He II absorption, leading to a provisional classification of B0:, with the uncertainty reflecting the problems of nebular contamination and the potential contribution from the nearby object.

The He II absorption at λ4542 and λ4686 appears consistent with the systemic radial velocity of 30 Dor, i.e., the object does not appear to be a candidate runaway in the radial direction. Indeed, as noted by Campbell et al. (2010), the bow shock is orientated towards R136 (and Brey 75/BAT99-100), suggesting it might well be related to an ionization front (with associated triggered star-formation) rather than a dynamical shock.

VFTS 500: This object appears as a single source in the HAWK-I imaging, as shown in Fig. 14 (in which the slight northern extension of the star is an artefact of the dither pattern used for the observations)⁹; the FLAMES spectroscopy reveals it as a double-lined binary. Observed in Medusa Field A, the first LR03 epoch was obtained on the same night as the first three LR02 spectra. The co-added spectrum from these epochs is shown in Fig. 12, displaying twin components in the He I and II lines. The nebular contamination thwarts precise classification, but both of the He I λ4026 components are greater in intensity than the He II λ4200 features, requiring a type of later than O6. Similarly, the intensity of the He II λ4200 profiles suggests a type of earlier than B0 for both components.

VFTS 586: The combined blue-region spectrum is shown in Fig. 12. The spectrum is classified O4 V((n)))((fc))z, in which the “c” suffix refers to C III emission at λλ4647-50-52 (Walborn et al. 2010).

VFTS 631: Inspection of the spectra of VFTS 631 reveals a single-lined binary, with radial velocity shifts of the order of 70 km   s^-1 in the final LR02 observation compared to the first epoch. To estimate the spectral type, the first three LR02 epochs (taken on the same night) were combined with the LR03 data, as shown in Fig. 12. The combined spectrum is classified as O9.7 III(n), in good agreement with published types of O9-B0 II (Walborn & Blades 1997) and O9.5 II (Bosch et al. 1999).

There are two obvious components to the nebular emission in the LR02 and LR03 spectra (e.g. twin-peaked emission in the [O III] lines). One of the components is consistent with the typical systemic velocity of 30 Dor (~270–280 km   s^-1, with the other blueshifted by approximately 50–60 km   s^-1). These are most likely from separate components of gas emission, but could also be indicative of a wind-blown bubble around the binary, depending on the systemic velocity of the system. A third (weaker), longer-wavelength nebular component is also visible in the [N II] and [S II] lines in the HR15N spectra.

VFTS 702: As with VFTS 631, radial velocity shifts are seen between the LR02 observations. To estimate the spectral type we combined the second and third LR02 epochs with the LR03 observations (all obtained on the same night), shown in Fig. 12 and classified as O8.5 V.

VFTS 751: An apparently single star, with the combined spectrum shown in Fig. 12. The stellar lines appear broadened by rotation, leading to a classification of O7-8 Vnnz.

5.1. Discussion

All of the spectra in Fig. 12 (except VFTS 178) contain some degree of nebular contribution; this is not surprising given the spatial extent of the H II region in 30 Dor. However, the most intense [O III] emission is seen in VFTS 410, 464, and 702, the three sources indentified by GC09 as clear YSO candidates. Both VFTS 410 and 464 are located immediately to the west of R136, while VFTS 702 is 2′ to the northeast. Echoing the discussion from Campbell et al. (2010) regarding the location of the candidate high-mass YSOs (which included VFTS 464), all three objects are in regions associated with molecular gas (Werner et al. 1978; Johansson et al. 1998) that comprise the second generation of star-formation around R136 (Walborn & Blades 1997; Walborn et al. 1999a, 2002).

Closer scrutiny of the correlation between Hα emission (from archival HST images) and the GC09 sources was provided by Vaidya et al. (2009), in which they discussed four of the VFTS sources: VFTS 464 was identified as a type II YSO (cf. the criteria from Chen et al. 2009) in a bright-rimmed dust pillar, and VFTS 320, 682 and 702 were each classifed as type III YSOs in H II regions. Vaidya et al. noted each of the latter three as comprising multiple sources in the Spitzer PSF. The close companions to VFTS 320 have already been noted above, while there are no obvious nearby companions to VFTS 702 in the HAWK-I images, nor in the IRSF catalogue. There is a faint IRSF source approximately 2′′ from VFTS 682: 05385578 − 6904285, with H  =  17.56  ±  0.09 mag and no detections in J and K_s. Intriguingly, VFTS 682 is only  ~ 175 from VFTS 702, both of which appear to be co-located with the most intense CO-emission in the region (see Fig. 1 from Johansson et al. 1998). This poses additional evolutionary questions regarding VFTS 682 (Bestenlehner et al., in prep.).

From their multi-band approach, Vaidya et al. (2009) reported VFTS 016 and 631 as non-YSOs. They describe VFTS 016 (GC09 053708.79 − 690720.3) as a bright star, consistent with the mid-IR excess perhaps arising from a bow shock. However, they flag VFTS 631 (GC09 053848.86 − 690828.0) as a galaxy. Notwithstanding the discussion of the spectroscopy of VFTS 631, the combined J- and K_s-band HAWK-I image appears unremarkable (Fig. 15), nor is it obviously extended in the WFI images.

Analysis of the SAGE data by Whitney et al. (2008) identified 1197 candidate YSOs using different selection criteria to GC09. The relative merits of the selection criteria were discussed by GC09, who note that 72.5% of their “probable” YSOs were not in the Whitney et al. catalogue owing to conservative identification criteria. While the nine probable YSOs from GC09 with VFTS spectroscopy in Table 4 are all interesting objects in the context of their mid-IR behaviour, none appear to be genuine high-mass YSOs. However, two of the more secure candidates (VFTS 410 and 464) appear to be genuine high-mass YSOs, and are not included in the Whitney et al. catalogue. Indeed, there is only one matched source between the VFTS targets and those from Whitney et al.: VFTS 178.

One potential explanation for the photometric properties of some of the (non-YSO) stars in Table 4 is that they are LMC analogues of the “dusty” B-type stars with 24 μm excesses discovered by Bolatto et al. (2007) in the SMC. They suggested that these might be caused by remnant accretion discs, planetary debris discs, or hot spots in the local interstellar cirrus; similar excesses for SMC stars have also been found by Ita et al. (2010) and Bonanos et al. (2010). However, Bolatto et al. (2007) report K_s  −  [24] colours of  ~ 6.5 mag. In contrast, the sources with 24 μm detections in Table 4 have larger K_s  −  [24] colours.

A range of physical explanations likely lie behind the mid-IR excesses in the sources listed in Table 4. One plausible explanation for some of the otherwise normal O- and early B-type stars could be the dusty shocks seen to be associated with massive stars in the Carina nebula (Smith et al. 2010). At the distance of the LMC these diffuse emission regions, depending on their spatial scales, can sometimes be within the resolution of the Spitzer observations.

Also included in Table 4 are four VFTS targets that GC09 classified as stellar sources with mid-IR excesses, VFTS 152 (classified here as B2 V), VFTS 323 (A5 II), VFTS 641 (B0.5: I:), and VFTS 842 (Be). VFTS 641 (aka Mk 11 Melnick 1985) is particularly bright at 24 μm ([24]  =  0.06  ±  0.12 mag from GC09), but we note that VFTS 655 is nearby (~8′′, cf. the angular resolution at 24 μm of 6′′) and is classified as late G/early K (Table  3), so confusion/crowding might be a factor. We only have UVES spectroscopy of VFTS 641, so the uncertainty in its classification arises from the lack of observations around 4100 Å; nevertheless this agrees well with previous classifications: B0.5 Ia (Melnick 1985); B0-0.5 Ia (Walborn 1986); B0 Ib (Walborn & Blades 1997); B0.2 IIII (Bosch et al. 1999).

6. Summary

We have introduced the VLT-FLAMES Tarantula Survey, which has provided high-quality spectroscopy of over 800 massive stars in the 30 Doradus region of the LMC. The survey targets are presented in Table 5, including B- and V-band optical photometry, and extensive cross-references with previous identifications in this well-studied region. Near-IR (JHK_s) magnitudes for our targets from the IRSF catalogue by Kato et al. (2007) are given in Table 6.

Spectral classifications are given for the massive emission-line stars in Sect. 4.1, including the discovery of a new WN5h star (VFTS 682) 2′ from R136, and a new B[e]-type star (VFTS 1003) just 85 to the west of R136. Classifications are also given for the cool-type stars observed by the survey that have radial velocities consistent with them being members of the LMC.

In Sect. 5 we investigated the spectral properties of 12 stars identified as definite or probable YSOs by GC09, finding an eclectic mixture of objects. VFTS 410 and 464 appear to be embedded massive stars, (i.e. bona fide high-mass YSOs), but others include the runaway star (VFTS 016), the new W-R star (VFTS 682), and three massive binaries. There are four additional stars in the survey with mid-IR exceses flagged by GC09, one of which is VFTS 641 (Mk 11), an early B-type supergiant that is particularly bright at 24 μm.

Analysis of the stellar radial velocities and multiplicity of the high-mass stars in the survey will be presented by Sana et al. and Dunstall et al. (both in prep.), but the power of the multi-epoch strategy has already been illustrated by two recent results. Namely, the discovery of VFTS 527 (R139) as an intriguing OIf + OIfc binary system (Taylor et al. 2011), and the lack of a detected companion to VFTS 016, which suggest that its peculiar radial velocity is caused by it being a massive runaway from the core of 30 Dor (Evans et al. 2010).

Analysis of the spectroscopy will be complemented in the future by two additional sources of photometry. Monitoring of selected fields across 30 Dor is part of an ongoing variability campaign with the 2 m Faulkes Telescope South at the Siding Spring Observatory, Australia. This builds on previous (and still ongoing) follow-up of the FLAMES fields from Evans et al. (2006) to determine periods for notable binaries (e.g. Ritchie et al., in prep.). The 30 Dor region was also one of the first fields observed by the ongoing VISTA Magellanic Clouds Survey (Cioni et al. 2011), with time-linked K_s-band observations to investigate variability.

The Tarantula Survey is by far the largest homogeneous spectroscopic study of extragalactic early-type stars undertaken to date. The 30 Dor region is the only “super star-cluster” at a well-known distance in which individual objects can be resolved spatially in optical light. This makes it the perfect target for the comprehensive studies required to address some of the fundamental questions that remain in our understanding of massive-star evolution, relying on the analysis of a statistically-significant and unbiased sample. The results of these studies will be presented in a series of forthcoming papers.

Online material

Appendix A: Observational epochs

---

Acknowledgments

Based on observations at the European Southern Observatory Very Large Telescope in programme 182.D-0222. We are indebted to Claudio Melo and the other ESO staff who have provided invaluable assistance with this programme, and to Brian Skiff for his dedicated and careful refinements of the published source catalogues. We are also grateful to Yuri Beletsky for his astrometric calibration, to Simone Zaggia for his advice on the WFI data, to Ian Hunter for his input to the original proposal, and to the referee for their helpful comments. STScI is operated by the Association of Universities for Research in Astronomy, under NASA contract NAS 5-26555, and S.d.M. acknowledges NASA Hubble Fellowship grant HST-HF-51270.01-A awarded by STScI. M.G. acknowledges the Royal Society for financial support. A.H. and S.S.D. acknowledge funding from the Spanish MICINN (AYA2008-06166-03-01, AYA2010-21697-C05-04, Consolider Ingenio 2010 CSD2006-70) and Gobierno de Canarias (ProID2010119). J.S.V. and P.R.D. acknowledge support from the Science and Technology Facilities Council. N.M. acknowledges the Bulgarian National Science Fund under grant DO02-85. V.H.B. acknowledges support from the Scottish Universities Physics Alliance (SUPA) and the Natural Sciences and Engineering Research Council of Canada (NSERC). J.M.A. acknowledges support from the Spanish Government through grants AYA2010-15081 and AYA2010-17631 and the Junta de Andalucía through grant P08-TIC-4075. A.Z.B. acknowledges research and travel support from the European Commission FP7 under Marie Curie International Re-integration Grant PIRG04-GA-2008-239335.

References

All Tables

All Figures